Classification of ADNI PET Images via Regularized 3D Functional Data Analysis

Area of Science:

• Neuroimaging Analysis
• Functional Data Analysis
• Machine Learning

Background:

• Accurate classification of 3D brain images is crucial for understanding neurological disorders.
• Existing machine learning methods may not fully capture spatial correlations within voxel-level imaging data.

Purpose of the Study:

• To develop and validate a novel penalized Haar wavelet approach for 3D brain image classification.
• To compare the performance of the proposed method against established machine learning techniques.
• To apply the method to Alzheimer's disease classification using PET imaging data.

Main Methods:

• A penalized Haar wavelet approach is proposed within the functional data analysis framework.
• Each 3D brain image is treated as a single functional input to account for spatial correlations.
• Extensive simulations were conducted for validation and comparison.

Main Results:

• The proposed penalized Haar wavelet method demonstrated superior classification accuracy compared to other commonly used machine learning methods.
• The method effectively identified relevant voxels contributing to classification.
• The approach showed practical advantages when applied to Alzheimer's disease classification.

Conclusions:

• The penalized Haar wavelet approach offers a robust and effective method for 3D brain image classification.
• This technique enhances both classification performance and the interpretability of results by identifying important brain regions.
• The method holds significant potential for clinical applications, particularly in diagnosing and understanding neurodegenerative diseases like Alzheimer's.